6.2.1. Environment Setup

Prerequisites and Core Dependencies

Your journey with PyRIT begins with ensuring your local system meets the necessary requirements. These are standard for modern Python development but are critical to verify before proceeding.

The most crucial prerequisite is access to an AI model endpoint. PyRIT is a framework for *testing* models, so it needs a target to interact with. For most examples in this section, we will assume you have access to an Azure OpenAI service endpoint.

Installation in an Isolated Environment

Always install tools like PyRIT within a dedicated virtual environment. This practice prevents dependency conflicts with other projects and ensures a clean, reproducible setup.

First, create and activate the virtual environment from your terminal:

# 1. Create a directory for your project
mkdir pyrit-red-team && cd pyrit-red-team

# 2. Create a virtual environment named 'venv'
python -m venv venv

# 3. Activate the environment
# On Windows:
# venvScriptsactivate
# On macOS/Linux:
source venv/bin/activate

Once your virtual environment is active (you should see (venv) in your terminal prompt), you can install PyRIT using pip:

# Install the core PyRIT package
pip install pyrit

PyRIT uses a modular design with “extras” for specific functionalities. For example, if you plan to use Azure Machine Learning orchestrators, you would install it with the `azureml` extra:

# Example: Installing with support for AzureML
pip install pyrit[azureml]

Configuring Secrets and Endpoints

Hardcoding API keys and endpoints directly into your scripts is a significant security risk. PyRIT is designed to load these sensitive values from a .env file in your project’s root directory. This file should be added to your .gitignore to prevent it from being committed to version control.

Create a file named .env in your project directory. The variables you define here will configure PyRIT’s connection to your target AI model. The following is an example configuration for connecting to an Azure OpenAI GPT-4 deployment.

# .env file for configuring PyRIT
# This file stores secrets and should NOT be committed to git.

# -- Azure OpenAI Chat Target Configuration --
# Your Azure OpenAI resource endpoint
AZURE_OPENAI_CHAT_ENDPOINT="https://your-resource-name.openai.azure.com/"

# The name of your chat model deployment (e.g., gpt-4, gpt-35-turbo)
AZURE_OPENAI_CHAT_DEPLOYMENT="your-deployment-name"

# Your Azure OpenAI API Key
API_KEY="your-azure-openai-api-key"

# -- Optional: Azure Content Safety --
# If you are using Azure Content Safety, provide its endpoint and key
AZURE_CONTENT_SAFETY_ENDPOINT="https://your-contentsafety-resource.cognitiveservices.azure.com/"
AZURE_CONTENT_SAFETY_KEY="your-content-safety-api-key"

By placing this file in your project root, PyRIT’s components, like the AzureOpenAIChatTarget, will automatically load and use these credentials, keeping your testing scripts clean and secure.

Verifying the Setup

With the installation and configuration complete, the final step is to verify that PyRIT can successfully communicate with your AI target. A simple script can confirm that your endpoint, deployment name, and API key are all correct.

Create a Python file named verify_setup.py and add the following code:

# verify_setup.py
# A simple script to confirm PyRIT can connect to the target.

import os
from pyrit.common import pyrit_target_init
from pyrit.prompt_target import AzureOpenAIChatTarget
from pyrit.prompt_normalizer import PromptRequestResponse, PromptRequestPiece

# Initialize the target. This will automatically load credentials from your .env file.
pyrit_target_init()
azure_chat_target = AzureOpenAIChatTarget()

# Create a simple, benign prompt request
request = PromptRequestResponse(
    [
        PromptRequestPiece(
            role="user",
            original_value="This is a test. If you see this, the connection is working.",
        )
    ]
)

# Send the request to the target
response = azure_chat_target.send_prompt(prompt_request=request)

# Print the response to confirm success
print("Connection successful! Model response:")
print(response.request_pieces[0].converted_value)

Run this script from your activated virtual environment:

python verify_setup.py

If your setup is correct, you will see a “Connection successful!” message followed by a response from the AI model. If you encounter an authentication error, double-check the values in your .env file. A connection timeout or name resolution error may indicate a problem with the endpoint URL or a network issue. With this verification complete, your PyRIT environment is now ready for orchestrating red team operations.